LAUSR

Computational imaging based on compressed sensing (CS) has shown potential for outperforming conventional techniques in many applications, but challenges arise when translating CS theory to practical imaging systems. Here we examine such challenges in two physical architectures under coherent and incoherent illumination. We describe hardware alignment protocols that can be used to optimize system performance for each case. We found that an architecture using coded masks located at a conjugate image plane outperformed an identical architecture using masks at a Fourier plane, enabling recovery of images with up to 64 times more resolvable points than pixels in the image sensor. We demonstrate and explain the basis for the tradeoff between achievable resolution and dynamic range of reconstructed CS images. Finally, we demonstrate that these principles can be applied beyond binary test targets by reconstructing a 480 × 480 image of a human tissue section from a 120 × 120 pixel sensor. These results provide a basis to further develop compressive imaging architectures for biomedical imaging and we also anticipate that these findings may be useful to investigators focused on translating CS theory to other real-world imaging applications.